Get the sleep current under spec
The pain. Change a peripheral clock gate, reflash, measure, read the number, change it again. Getting a microcontroller under its standby budget is hours of that loop, and the loop needs a human in it only because the instrument cannot be asked a question.
What you do. Power the board through the analyser and have its firmware toggle a marker line around each power state. Your agent edits and flashes locally with its own toolchain, then arms a capture and reads the segments back over the API.
What you get. Current per power state, not one averaged number that hides which state is expensive. And the loop runs without you sitting in it.
Catch the transient you would otherwise miss
The pain. The sleep floor is microamps and the radio burst is amps. One range cannot see both, so you either miss the floor or clip the burst — and averaging erases the spike that ruined the battery estimate in the first place.
What you do. One capture. Autoranging happens per sample, recording is lossless, and it runs for hours rather than seconds.
What you get. The microamp floor and the amp burst in the same recording, with the minimum and maximum of every interval preserved — so an event a mean would have flattened is still there when you zoom in.
Make power a regression test
The pain. Nobody notices the commit that cost 200 µA. It ships, and months later the battery numbers come back wrong and the bisect starts from a hundred commits.
What you do. The same capture, scripted through the local API, on every build. The API makes this possible today; you write the harness that fits your pipeline.
What you get. The power budget checked like any other test, so a regression is caught on the day it lands rather than the quarter it ships.